[Qemu-devel] [PATCH v2 1/1] spapr: Set compat type for CPUs of a core

[Qemu-devel] [PATCH v2 1/1] spapr: Set compat type for CPUs of a core

[Bharata B Rao]

Compat CPU type is typically specified on -cpu cmdline option like:
-cpu host,compat=power7 or -cpu POWER8E,compat=power7 etc. When
compat is specified on -cpu cmdline, apply the same to all the
CPUs that get created as part of CPU core devices.

This patch takes care of compat property that is specified with
-cpu cmdline option only. The other way to specify this property
is via -global cmdline option and that usage isn't addressed by
this patch because -global is already broken for some CPU class
in PowerPC. There are two issues with using -global on CPU class.

- We specify alias names for CPUs commonly and use of alias names
  won't work with -global seamlessly. For eg, When "-cpu host" is
  specified, the actual CPU type that gets created
  is host-powerpc64-cpu. Hence specifying -global host.compat=power7
  will not work, so it has to be -global host-powerpc64-cpu.compat=power7.

- PowerPC class names have . (dot) and opts parsing doesn't like it.
  Specifying -global POWER8E_v2.1-powerpc64-cpu.compat=power7 will not
  work as the driver name gets extracted as POWER8E_v2 and hence setting
  of global property fails.

The above two aspects could be considered/fixed separately from this
patch as this patch allows existing uses of -cpu cpuname,compat= to
work correctly after the introducton of sPAPR CPU cores.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
---
Changes in v2:
- No need for a separate property named compat for cores.
- Simplified spapr_get_cpu_compat_type() based on David's review.

v1: https://lists.gnu.org/archive/html/qemu-devel/2016-06/msg06279.html

 hw/ppc/spapr_cpu_core.c | 32 ++++++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)

diff --git a/hw/ppc/spapr_cpu_core.c b/hw/ppc/spapr_cpu_core.c
index 3a5da09..e8873fd 100644
--- a/hw/ppc/spapr_cpu_core.c
+++ b/hw/ppc/spapr_cpu_core.c
@@ -96,6 +96,28 @@ char *spapr_get_cpu_core_type(const char *model)
     return core_type;
 }
 
+/*
+ * Returns the CPU compat type specified in -cpu @model.
+ */
+static char *spapr_get_cpu_compat_type(const char *model)
+{
+    char *model_str = g_strdup(model);
+    char *featurestr, *compat = NULL;
+
+    featurestr = model_str ? strtok(model_str, ",") : NULL;
+    while (featurestr) {
+        if (!strncmp(featurestr, "compat=", 7)) {
+            compat = g_strdup(featurestr + 7);
+            goto out;
+        }
+        featurestr = strtok(NULL, ",");
+    }
+
+out:
+    g_free(model_str);
+    return compat;
+}
+
 static void spapr_core_release(DeviceState *dev, void *opaque)
 {
     sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
@@ -285,6 +307,8 @@ static void spapr_cpu_core_realize(DeviceState *dev, Error **errp)
     CPUCore *cc = CPU_CORE(OBJECT(dev));
     const char *typename = object_class_get_name(sc->cpu_class);
     size_t size = object_type_get_instance_size(typename);
+    MachineState *machine = MACHINE(qdev_get_machine());
+    char *compat = spapr_get_cpu_compat_type(machine->cpu_model);
     Error *local_err = NULL;
     Object *obj;
     int i;
@@ -300,11 +324,18 @@ static void spapr_cpu_core_realize(DeviceState *dev, Error **errp)
         if (local_err) {
             goto err;
         }
+        if (compat) {
+            object_property_set_str(obj, compat, "compat", &local_err);
+            if (local_err) {
+                goto err;
+            }
+        }
     }
     object_child_foreach(OBJECT(dev), spapr_cpu_core_realize_child, &local_err);
     if (local_err) {
         goto err;
     } else {
+        g_free(compat);
         return;
     }
 
@@ -315,6 +346,7 @@ err:
         i--;
     }
     g_free(sc->threads);
+    g_free(compat);
     error_propagate(errp, local_err);
 }
 
-- 
2.1.0

Re: [Qemu-devel] [PATCH v2 1/1] spapr: Set compat type for CPUs of a core

[David Gibson]

[-- Attachment #1: Type: text/plain, Size: 4625 bytes --]

On Mon, Jun 27, 2016 at 11:49:46AM +0530, Bharata B Rao wrote:
> Compat CPU type is typically specified on -cpu cmdline option like:
> -cpu host,compat=power7 or -cpu POWER8E,compat=power7 etc. When
> compat is specified on -cpu cmdline, apply the same to all the
> CPUs that get created as part of CPU core devices.
> 
> This patch takes care of compat property that is specified with
> -cpu cmdline option only. The other way to specify this property
> is via -global cmdline option and that usage isn't addressed by
> this patch because -global is already broken for some CPU class
> in PowerPC. There are two issues with using -global on CPU class.
> 
> - We specify alias names for CPUs commonly and use of alias names
>   won't work with -global seamlessly. For eg, When "-cpu host" is
>   specified, the actual CPU type that gets created
>   is host-powerpc64-cpu. Hence specifying -global host.compat=power7
>   will not work, so it has to be -global host-powerpc64-cpu.compat=power7.
> 
> - PowerPC class names have . (dot) and opts parsing doesn't like it.
>   Specifying -global POWER8E_v2.1-powerpc64-cpu.compat=power7 will not
>   work as the driver name gets extracted as POWER8E_v2 and hence setting
>   of global property fails.
> 
> The above two aspects could be considered/fixed separately from this
> patch as this patch allows existing uses of -cpu cpuname,compat= to
> work correctly after the introducton of sPAPR CPU cores.
> 
> Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
> ---
> Changes in v2:
> - No need for a separate property named compat for cores.
> - Simplified spapr_get_cpu_compat_type() based on David's review.

I'm still not seeing why this is actually necessary.  If the compat
property is specified with -global, it will automatically be
propagated to all threads without core intervention, yes?

So shouldn't we just make the -cpu fallback effectively set the same
-global property, rather than bouncing it through the core object?

> 
> v1: https://lists.gnu.org/archive/html/qemu-devel/2016-06/msg06279.html
> 
>  hw/ppc/spapr_cpu_core.c | 32 ++++++++++++++++++++++++++++++++
>  1 file changed, 32 insertions(+)
> 
> diff --git a/hw/ppc/spapr_cpu_core.c b/hw/ppc/spapr_cpu_core.c
> index 3a5da09..e8873fd 100644
> --- a/hw/ppc/spapr_cpu_core.c
> +++ b/hw/ppc/spapr_cpu_core.c
> @@ -96,6 +96,28 @@ char *spapr_get_cpu_core_type(const char *model)
>      return core_type;
>  }
>  
> +/*
> + * Returns the CPU compat type specified in -cpu @model.
> + */
> +static char *spapr_get_cpu_compat_type(const char *model)
> +{
> +    char *model_str = g_strdup(model);
> +    char *featurestr, *compat = NULL;
> +
> +    featurestr = model_str ? strtok(model_str, ",") : NULL;
> +    while (featurestr) {
> +        if (!strncmp(featurestr, "compat=", 7)) {
> +            compat = g_strdup(featurestr + 7);
> +            goto out;
> +        }
> +        featurestr = strtok(NULL, ",");
> +    }
> +
> +out:
> +    g_free(model_str);
> +    return compat;
> +}
> +
>  static void spapr_core_release(DeviceState *dev, void *opaque)
>  {
>      sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
> @@ -285,6 +307,8 @@ static void spapr_cpu_core_realize(DeviceState *dev, Error **errp)
>      CPUCore *cc = CPU_CORE(OBJECT(dev));
>      const char *typename = object_class_get_name(sc->cpu_class);
>      size_t size = object_type_get_instance_size(typename);
> +    MachineState *machine = MACHINE(qdev_get_machine());
> +    char *compat = spapr_get_cpu_compat_type(machine->cpu_model);
>      Error *local_err = NULL;
>      Object *obj;
>      int i;
> @@ -300,11 +324,18 @@ static void spapr_cpu_core_realize(DeviceState *dev, Error **errp)
>          if (local_err) {
>              goto err;
>          }
> +        if (compat) {
> +            object_property_set_str(obj, compat, "compat", &local_err);
> +            if (local_err) {
> +                goto err;
> +            }
> +        }
>      }
>      object_child_foreach(OBJECT(dev), spapr_cpu_core_realize_child, &local_err);
>      if (local_err) {
>          goto err;
>      } else {
> +        g_free(compat);
>          return;
>      }
>  
> @@ -315,6 +346,7 @@ err:
>          i--;
>      }
>      g_free(sc->threads);
> +    g_free(compat);
>      error_propagate(errp, local_err);
>  }
>  

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [Qemu-devel] [PATCH v2 1/1] spapr: Set compat type for CPUs of a core

[Bharata B Rao]

On Mon, Jun 27, 2016 at 04:51:34PM +1000, David Gibson wrote:
> On Mon, Jun 27, 2016 at 11:49:46AM +0530, Bharata B Rao wrote:
> > Compat CPU type is typically specified on -cpu cmdline option like:
> > -cpu host,compat=power7 or -cpu POWER8E,compat=power7 etc. When
> > compat is specified on -cpu cmdline, apply the same to all the
> > CPUs that get created as part of CPU core devices.
> > 
> > This patch takes care of compat property that is specified with
> > -cpu cmdline option only. The other way to specify this property
> > is via -global cmdline option and that usage isn't addressed by
> > this patch because -global is already broken for some CPU class
> > in PowerPC. There are two issues with using -global on CPU class.
> > 
> > - We specify alias names for CPUs commonly and use of alias names
> >   won't work with -global seamlessly. For eg, When "-cpu host" is
> >   specified, the actual CPU type that gets created
> >   is host-powerpc64-cpu. Hence specifying -global host.compat=power7
> >   will not work, so it has to be -global host-powerpc64-cpu.compat=power7.
> > 
> > - PowerPC class names have . (dot) and opts parsing doesn't like it.
> >   Specifying -global POWER8E_v2.1-powerpc64-cpu.compat=power7 will not
> >   work as the driver name gets extracted as POWER8E_v2 and hence setting
> >   of global property fails.
> > 
> > The above two aspects could be considered/fixed separately from this
> > patch as this patch allows existing uses of -cpu cpuname,compat= to
> > work correctly after the introducton of sPAPR CPU cores.
> > 
> > Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
> > ---
> > Changes in v2:
> > - No need for a separate property named compat for cores.
> > - Simplified spapr_get_cpu_compat_type() based on David's review.
> 
> I'm still not seeing why this is actually necessary.  If the compat
> property is specified with -global, it will automatically be
> propagated to all threads without core intervention, yes?
> 
> So shouldn't we just make the -cpu fallback effectively set the same
> -global property, rather than bouncing it through the core object?

Yes and the below patch against Igor's cpu_parse_into_global_props_V2
branch exactly does that for sPAPR.

===========
spapr: Parse CPU feature only once

From: Bharata B Rao <bharata@linux.vnet.ibm.com>

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
---
 hw/ppc/spapr.c |   22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index 0636642..be24a37 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -1727,6 +1727,10 @@ static void ppc_spapr_init(MachineState *machine)
     long load_limit, fw_size;
     bool kernel_le = false;
     char *filename;
+    CPUClass *cc;
+    ObjectClass *oc;
+    const char *typename;
+    gchar **model_pieces;
 
     msi_nonbroken = true;
 
@@ -1785,6 +1789,24 @@ static void ppc_spapr_init(MachineState *machine)
     if (machine->cpu_model == NULL) {
         machine->cpu_model = kvm_enabled() ? "host" : "POWER7";
     }
+
+    model_pieces = g_strsplit(machine->cpu_model, ",", 2);
+    if (!model_pieces[0]) {
+        error_report("Invalid/empty CPU model name");
+        exit(1);
+    }
+
+    oc = cpu_class_by_name(TYPE_POWERPC_CPU, model_pieces[0]);
+    if (oc == NULL) {
+        error_report("Unable to find CPU definition: %s", model_pieces[0]);
+        exit(1);
+    }
+
+    typename = object_class_get_name(oc);
+    cc = CPU_CLASS(oc);
+    cc->parse_features(typename, model_pieces[1], &error_fatal);
+    g_strfreev(model_pieces);
+
     for (i = 0; i < smp_cpus; i++) {
         cpu = cpu_ppc_init(machine->cpu_model);
         if (cpu == NULL) {

===========

So essentially this uses cc->parse_features() to parse the cpu feature string
even before any CPU is initialized and cpu_common_parse_features() will
ensure that compat= and any other properties that are part of -cpu will
be set globally only once. I am following what Igor did for x86 here.

Igor's tree doesn't yet contain PowerPC hotplug patches, hence I couldn't
verify this with spapr cpu cores, but I believe that this is all that is
required (on top of Igor's cpu features work) to ensure that compat= is
applied automatically to all the created CPU threads.

However as I note in the patch description of the original patch in this
mail, -global is broken for certain cpu type names.

-cpu host -global host-powerpc64-cpu.compat=power7
-cpu POWER8 -global POWER8-powerpc64-cpu.compat=power7

work, but

-cpu POWER8E -global POWER8E_v2.1-powerpc64-cpu.compat=power7

doesn't since cpu typename has . (dot) and this needs to be fixed.

Regards,
Bharata.

Re: [Qemu-devel] [PATCH v2 1/1] spapr: Set compat type for CPUs of a core

[Eduardo Habkost]

On Mon, Jun 27, 2016 at 02:10:48PM +0530, Bharata B Rao wrote:
[...]
> However as I note in the patch description of the original patch in this
> mail, -global is broken for certain cpu type names.
> 
> -cpu host -global host-powerpc64-cpu.compat=power7
> -cpu POWER8 -global POWER8-powerpc64-cpu.compat=power7
> 
> work, but
> 
> -cpu POWER8E -global POWER8E_v2.1-powerpc64-cpu.compat=power7
> 
> doesn't since cpu typename has . (dot) and this needs to be fixed.

-global accepts the saner QemuOpts syntax, too. So this should
work:

  -global driver=POWER8E_v2.1-powerpc64-cpu,property=compat,value=power7

-- 
Eduardo